汽车工程 ›› 2021, Vol. 43 ›› Issue (11): 1683-1692.doi: 10.19562/j.chinasae.qcgc.2021.11.014

• • 上一篇    下一篇

考虑多维驾驶特性的制动反应时间预测模型

郭柏苍1,谢宪毅1,金立生1,戎辉2(),贺阳1,纪丙东1   

  1. 1.燕山大学车辆与能源学院,秦皇岛  066004
    2.中国汽车技术研究中心有限公司,天津  300300
  • 收稿日期:2021-05-24 修回日期:2021-08-15 出版日期:2021-11-25 发布日期:2021-11-22
  • 通讯作者: 戎辉 E-mail:ronghui@catarc.ac.cn
  • 基金资助:
    国家重点研发计划(2018YFB1600501);国家自然科学基金(52072333);河北省省级科技计划(E2020203092);河北省省级科技计划(20310801D);教育部2020年第一批产学合作协同育人项目(202001SJ05)

Braking Response Time Prediction Model Based on Multi-dimensional Driving Characteristics

Baicang Guo1,Xianyi Xie1,Lisheng Jin1,Hui Rong2(),Yang He1,Bingdong Ji1   

  1. 1.School of Vehicle and Energy,Yanshan University,Qinhuangdao  066004
    2.China Automotive Technology and Research Center Co. ,Ltd. ,Tianjin  300300
  • Received:2021-05-24 Revised:2021-08-15 Online:2021-11-25 Published:2021-11-22
  • Contact: Hui Rong E-mail:ronghui@catarc.ac.cn

摘要:

为了准确预测驾驶人的制动反应时间,建立了考虑差异化驾驶人特性的制动反应时间预测模型。以多种次任务驾驶行为作为差异化驾驶人特性的诱导因素设计了试验,在封闭的城市道路展开了实车试验并采集了制动反应时间数据,以自报式信息采集法获取了受试者的多维度驾驶特性变量数据,使用结构方程模型解构制动反应时间的影响因素并以路径系数优化BP神经网络权值,建立了基于SEM-BP神经网络的驾驶人制动反应时间预测模型。验证和测试结果表明,所提出的制动反应时间预测模型总体的回归R值大于0.9,总误差为0.032 4,有更好的预测精度和拟合性能,能够在考虑驾驶人多维度特性的同时降低网络收敛不稳定导致鲁棒性差的问题。

关键词: 交通工程, 汽车人因工程, 制动反应时间, 驾驶人特性, 结构方程模型, BP神经网络

Abstract:

In order to accurately predict the driver's braking reaction time (BRT), a BRT prediction model based on the characteristics of differentiated drivers is built. The experiment is designed with multi task driving behavior as the inducing factors of differentiated driver characteristics, then the real vehicle experiment is carried out on closed urban road and the BRT data is collected. The drivers' multi-dimensional driving characteristics variable data is obtained by the self-reported information collection method. The structural equation model (SEM) is used to deconstruct the influencing factors of BRT and the path coefficients are used to optimize the weights of back propagation neural network (BPNN). Finally, a prediction model of BRT based on SEM-BPNN is established. The verification and test results show that the overall regression R value of the proposed BRT prediction model is greater than 0.9 and the total error is 0.032 4. It has better prediction accuracy and fitting performance, moreover, it can reduce the problem of poor robustness caused by unstable network convergence while considering the multi-dimensional characteristics of drivers.

Key words: traffic engineering, automotive human factors engineering, breaking reaction time, driver characteristics, structural equation model, BP neural network